This immediately reminds me of: Given a list of Tetris moves, return the number of completed lines
That code-golf question gathered 2 PHP answers, a C answer and a Ruby answer. No golfing languages appeared (although I for one would love to see a golfing language entry).
The main reason appears to be that it required multiple related but distinct stages: compression of piece shapes, simulation of game physics, and recognition of completed lines. This is what regular languages were built for, but golfing languages struggle to combine multiple parts; they're optimised for solving individual problems.
An important note is that the tetris challenge was coherent and fun. Just gluing challenges together to put golfing languages at a disadvantage is no fun for anybody. It was also flexible in its input and output format, and didn't require specialist functionality (e.g. there was no need to output images, and no need to handle interactivity).
Nothing about the challenge did anything to exclude or even discourage golfing languages; it simply didn't appeal to them.
Or: non-code golf challenges
The obvious one, as already pointed out by Fatalize, and emphasised by DLosc:
Golfing languages aren't as good at challenges that aren't code golf.
As a side-note, I really like seeing golfing language answers to any question. They often take a much higher-level approach to problems and there's a lot to learn from reading them. And I say this as somebody who answers here almost exclusively in C. Whoever "wins" the challenge doesn't matter one bit; it's just about having a target to aim for.